In courtrooms across the United States, judges are making sentencing decisions with the help of a score generated by a private company's algorithm. The software is called COMPAS. It assigns defendants a number between 1 and 10 representing their predicted likelihood of committing another crime — a number that influences bail, sentence lengths, and parole eligibility for hundreds of thousands of Americans every year. The formula is a trade secret. Defendants have no right to see it, challenge it, or know what factors it weighs.
In 2016, ProPublica analyzed COMPAS scores for over 10,000 defendants in Broward County, Florida. Black defendants were nearly twice as likely as white defendants to be incorrectly flagged as high risk when they did not go on to reoffend. White defendants who did reoffend were far more likely to have been labeled low risk. The errors ran consistently in one direction.
The company disputed the findings. Researchers responded that both statistical arguments could be simultaneously true — and that the way COMPAS defines fairness mathematically guarantees the racial disparity ProPublica found will exist. The debate has continued for years. The algorithm is still in use.
It uses factors like age, employment history, and prior criminal record — variables already skewed by decades of unequal policing. A Dartmouth study found it was no more accurate at predicting recidivism than random people recruited from the internet with no criminal justice training whatsoever.
After ProPublica's investigation, Broward County officials were asked whether they planned to change how they used COMPAS. They said no. The manufacturer had no comment. The scores continue to be generated, used, and shielded from the people they affect most.
